At the end of this part you have a checklist of what to clarify before you promise a customer a Fabric App, and a list of what our demo deliberately simplifies.
Time to read: 15 minutes.
Credits: none.
The demo is secure for signed-in colleagues who use the app as intended. Most gaps concern deliberate misuse, auditability and roles: exactly the questions a controller or an auditor will ask.
What the demo simplifies
The checklist for every customer project
Several of these can rule the approach out, so ask them before the offer, not after.
Platform
- The capacity region supports Fabric Apps
- Capacity or Pro/PPU licensing for app users is clarified
- Both tenant settings can be enabled, and for which groups
- The customer accepts a preview feature (no SLA, changes possible)
- AI tools are allowed in the customer’s environment, and it’s clear who pays for AI credits
Users and access
- Only internal users, or also externals (then Entra guests, or a different architecture such as Power BI Embedded)
- Roles: who plans what, who sees what, who approves
- RLS in the semantic model matches the app roles
- Who owns the semantic model and grants Build/Read
Planning process
- Scenarios and versions needed (Base/Best/Worst, forecast versions)
- Approval steps and who finalizes
- Is a full change history required (auditors, controlling)
- Planning unit (euros or k€, clearly labeled), currency, fiscal or calendar year, plausibility warnings
Operations
- Dev/test/prod workspaces and deployment process
- Repository location and access (private)
- Who maintains the app after go-live and updates Rayfin versions
- Reporting on saved plans (plan vs. actual in Power BI)
The model contract: the BI topic nobody mentions
This one matters most for us. The app depends on specific measures, tables, columns and keys in your semantic model. So: adding measures is safe; renaming or removing them, changing keys like RegionID, or publishing the model as a new item breaks the app or orphans saved forecasts.
What I’d do in every project:
- Keep a list of everything the app uses from the model (my „model contract“).
- Before publishing model changes, run a test query with npx fabric-app-data query.
- Treat model changes like an API change: announce, test, then publish.
A prompt for later: audit trail
If the customer needs a change history, this is my starting point. Note that the app writes these records itself, so deliberate misuse could bypass them; audit-proof history needs enforcement on the database side.
Add an audit trail, no changes to DAX or measures:
1. New append-only entity "ForecastChange" (id, scenarioId, regionId, productGroupId, month, oldAmount, newAmount, changedBy, changedByName, changedAt). Every save writes one record per changed value. Records can be read by all signed-in users, but never updated or deleted.
2. Add finalizedBy and finalizedAt to Scenario, set when a scenario becomes Final.
3. A "History" panel: for a selected grid cell, show all changes with old value, new value, user and time; for a scenario, show who created and finalized it.
Explain the database changes before deploying, then deploy with npx rayfin up and confirm the deployment in your summary.
Where Fabric Apps fit today
Good fits right now: internal tools next to existing reports, like forecast and plan input, mapping-table maintenance, small approval workflows. Not yet a fit: portals for many external users without Microsoft accounts, and anything that needs audit-proof server-side rules before Microsoft ships them.
⚠️ And just remember: don’t use preview features in production! I know you love the risk, but just NO! Until Fabric Apps is generally available, a customer project means a pilot or an internal prototype.
That’s the end of the main series. Thank you for building along. Tell me what you built with #HumanAICoProgramming.
Fabric Apps is in preview; menus and commands may change. Written October 2026. Co-authored with Claude as co-learner. Previous: Part 9 · Bonus: the lakehouse supplier portal. #HumanAICoProgramming